Hybrid Leveraging AI Models and Deep Learning Framework Using MRI for Brain Tumor Detection

Abstract

Tumours of the brain are among the most severe neurological diseases, and fast and accurate diagnosis is essential for increasing treatment effectiveness and patient survival. Magnetic Resonance Imaging (MRI) can provide detailed visualisation of cerebral anomalies, but manual interpretation is often labour intensive and prone to diagnostic inconsistency. The effectiveness of current machine learning and deep learning methods are promising, but still facing challenges such as tumour heterogeneity, feature representation, multi-scale localisation and computational efficiency as major obstacles. We propose a Hybrid Leveraging Artificial Intelligence and Deep Learning Framework for improving MRI based brain tumour detection and classification in this paper. The proposed system includes a modified architecture of YOLOv8 with GhostConv blocks, a Vision Transformer Encoder (VTE), an improved Spatial Pyramid Pooling Fast Plus (SPPF+) module, multi-scale feature fusion strategies, and RT-DETR detection heads. Transfer learning with pre-trained COCO weights is used to improve feature generalisation and reduce dependence on large annotated medical data sets. The MRI images are preprocessed and divided into training and testing sets. The model is trained to accurately localise and classify brain tumours of different severity levels.The proposed approach is evaluated experimentally on the Kaggle Brain Tumour MRI dataset and the results show its effectiveness. The model achieved an accuracy of 99.6%, precision of 99.5%, recall of 99.6% and F1-score of 99.5% outperforming several traditional and state-of-the-art methods such as DenseNet, CNN Ensemble, SVM and Xception-based models. Further support for the framework’s robustness is provided by the analysis of the confusion matrix, showing few false-positive and false-negative predictions.The results show that the hybrid architecture proposed provides excellent accuracy, reliability and computing efficiency in the detection and localisation of brain tumour from MRI data. The combination of sophisticated deep learning, transformer-based attention mechanisms and multi-scale feature fusion makes the system a promising computer-aided diagnostic instrument, which can assist radiologists in early diagnosis, clinical decision making and treatment planning.